Add bounded retry logic for lock publish quorum

This commit is contained in:
thePR0M3TH3AN committed 2026-02-14 19:08:53 -05:00
1 parent 3c5bf0ba9e
commit f22bcfb713
2 files changed
+239 -47

No files matched your search

+138 -23
View File
@@ -65,23 +65,155 @@ describe('lock-ops', () => {
});
describe('publishLock', () => {
it('succeeds with partial primary success when minimum successes is 1', async () => {
it('retries transient failures and succeeds on a later attempt', async () => {
const telemetry = [];
const sleepCalls = [];
let publishCalls = 0;
const mockPool = {
publish: (relays) => relays.map((relay) => (
relay.includes('ok') ? Promise.resolve('ok') : Promise.reject(new Error('offline'))
)),
publish: (relays) => {
publishCalls += 1;
if (publishCalls === 1) {
return relays.map(() => Promise.reject(new Error('connection reset by peer')));
}
return relays.map(() => Promise.resolve('ok'));
},
close: () => {},
};
const event = { id: 'evt-1' };
const result = await publishLock(['wss://ok-1', 'wss://bad-1'], event, {
const event = { id: 'evt-retry' };
const result = await publishLock(['wss://primary-a', 'wss://primary-b'], event, {
poolFactory: () => mockPool,
getPublishTimeoutMsFn: () => 200,
getMinSuccessfulRelayPublishesFn: () => 1,
getRelayFallbacksFn: () => [],
retryAttempts: 3,
retryBaseDelayMs: 500,
retryCapDelayMs: 8_000,
randomFn: () => 0.5,
sleepFn: async (ms) => {
sleepCalls.push(ms);
},
telemetryLogger: (line) => {
telemetry.push(JSON.parse(line));
},
});
assert.strictEqual(result, event);
assert.strictEqual(publishCalls, 2);
assert.deepStrictEqual(sleepCalls, [250]);
assert.strictEqual(telemetry.length, 2);
assert.deepStrictEqual(telemetry[0], {
event: 'lock_publish_retry',
attempt: 1,
relayUrl: 'wss://primary-a',
errorCategory: 'connection_reset',
elapsedMs: telemetry[0].elapsedMs,
nextDelayMs: 250,
});
assert.deepStrictEqual(telemetry[1], {
event: 'lock_publish_retry',
attempt: 1,
relayUrl: 'wss://primary-b',
errorCategory: 'connection_reset',
elapsedMs: telemetry[1].elapsedMs,
nextDelayMs: 250,
});
assert.ok(Number.isInteger(telemetry[0].elapsedMs));
assert.ok(Number.isInteger(telemetry[1].elapsedMs));
});
it('fails after retry budget is exhausted for persistent transient failures', async () => {
let publishCalls = 0;
const sleepCalls = [];
const mockPool = {
publish: (relays) => {
publishCalls += 1;
return relays.map(() => Promise.reject(new Error('relay unavailable')));
},
close: () => {},
};
await assert.rejects(
() => publishLock(['wss://primary-down'], { id: 'evt-exhausted' }, {
poolFactory: () => mockPool,
getPublishTimeoutMsFn: () => 200,
getMinSuccessfulRelayPublishesFn: () => 1,
getRelayFallbacksFn: () => [],
retryAttempts: 3,
randomFn: () => 1,
sleepFn: async (ms) => {
sleepCalls.push(ms);
},
telemetryLogger: () => {},
}),
/Failed relay publish quorum in publish phase: 0\/1 successful \(required=1, timeout=200ms, attempts=3\)/,
);
assert.strictEqual(publishCalls, 3);
assert.deepStrictEqual(sleepCalls, [500, 1000]);
});
it('fails without retry for persistent validation failures', async () => {
let publishCalls = 0;
const mockPool = {
publish: (relays) => {
publishCalls += 1;
return relays.map(() => Promise.reject(new Error('validation failed: invalid signature')));
},
close: () => {},
};
await assert.rejects(
() => publishLock(['wss://primary-bad'], { id: 'evt-permanent' }, {
poolFactory: () => mockPool,
getPublishTimeoutMsFn: () => 200,
getMinSuccessfulRelayPublishesFn: () => 1,
getRelayFallbacksFn: () => [],
retryAttempts: 4,
sleepFn: async () => {
throw new Error('should not sleep for permanent failures');
},
telemetryLogger: () => {
throw new Error('should not emit retry telemetry for permanent failures');
},
}),
/Failed relay publish quorum in publish phase: 0\/1 successful \(required=1, timeout=200ms, attempts=4\)/,
);
assert.strictEqual(publishCalls, 1);
});
it('returns success when mixed relay outcomes still satisfy quorum', async () => {
let publishCalls = 0;
const mockPool = {
publish: (relays) => {
publishCalls += 1;
return relays.map((relay) => (
relay === 'wss://relay-ok'
? Promise.resolve('ok')
: Promise.reject(new Error('validation failed: rejected event'))
));
},
close: () => {},
};
const event = { id: 'evt-mixed' };
const result = await publishLock(['wss://relay-ok', 'wss://relay-bad'], event, {
poolFactory: () => mockPool,
getPublishTimeoutMsFn: () => 200,
getMinSuccessfulRelayPublishesFn: () => 1,
getRelayFallbacksFn: () => [],
retryAttempts: 3,
sleepFn: async () => {
throw new Error('should not sleep when quorum is satisfied');
},
});
assert.strictEqual(result, event);
assert.strictEqual(publishCalls, 1);
});
it('uses fallback relays to satisfy min success quorum', async () => {
@@ -102,22 +234,5 @@ describe('lock-ops', () => {
assert.strictEqual(result, event);
});
it('fails when successful publishes stay below configured minimum', async () => {
const mockPool = {
publish: (relays) => relays.map(() => Promise.reject(new Error('offline'))),
close: () => {},
};
await assert.rejects(
() => publishLock(['wss://primary-bad'], { id: 'evt-3' }, {
poolFactory: () => mockPool,
getPublishTimeoutMsFn: () => 200,
getMinSuccessfulRelayPublishesFn: () => 2,
getRelayFallbacksFn: () => ['wss://fallback-bad'],
}),
/Failed relay publish quorum in publish phase: 0\/2 successful \(required=2, timeout=200ms\)/,
);
});
});
});